source: libcfa/src/concurrency/kernel.hfa @ 5db836e

//
// Cforall Version 1.0.0 Copyright (C) 2016 University of Waterloo
//
// The contents of this file are covered under the licence agreement in the
// file "LICENCE" distributed with Cforall.
//
// kernel --
//
// Author           : Thierry Delisle
// Created On       : Tue Jan 17 12:27:26 2017
// Last Modified By : Peter A. Buhr
// Last Modified On : Tue Feb  4 12:29:26 2020
// Update Count     : 22
//

#pragma once

#include <stdbool.h>
#include <stdint.h>

#include "invoke.h"
#include "time_t.hfa"
#include "coroutine.hfa"

#include "containers/stackLockFree.hfa"

extern "C" {
#include <pthread.h>
#include <semaphore.h>
}

//-----------------------------------------------------------------------------
// Locks
struct semaphore {
        __spinlock_t lock;
        int count;
        __queue_t($thread) waiting;
};

void  ?{}(semaphore & this, int count = 1);
void ^?{}(semaphore & this);
bool   P (semaphore & this);
bool   V (semaphore & this);
bool   V (semaphore & this, unsigned count);


//-----------------------------------------------------------------------------
// Processor
extern struct cluster * mainCluster;

// Processor id, required for scheduling threads
struct __processor_id_t {
        unsigned id;

        #if !defined(__CFA_NO_STATISTICS__)
                struct __stats_t * stats;
        #endif
};

coroutine processorCtx_t {
        struct processor * proc;
};

// Wrapper around kernel threads
struct __attribute__((aligned(128))) processor {
        // Main state
        inline __processor_id_t;

        // Cluster from which to get threads
        struct cluster * cltr;

        // Set to true to notify the processor should terminate
        volatile bool do_terminate;

        // Coroutine ctx who does keeps the state of the processor
        struct processorCtx_t runner;

        // Name of the processor
        const char * name;

        // Handle to pthreads
        pthread_t kernel_thread;

        // RunThread data
        // Action to do after a thread is ran
        $thread * destroyer;

        // Preemption data
        // Node which is added in the discrete event simulaiton
        struct alarm_node_t * preemption_alarm;

        // If true, a preemption was triggered in an unsafe region, the processor must preempt as soon as possible
        bool pending_preemption;

        // Idle lock (kernel semaphore)
        __bin_sem_t idle;

        // Termination synchronisation (user semaphore)
        semaphore terminated;

        // pthread Stack
        void * stack;

        // Link lists fields
        Link(processor) link;

        #if !defined(__CFA_NO_STATISTICS__)
                int print_stats;
                bool print_halts;
        #endif

#ifdef __CFA_DEBUG__
        // Last function to enable preemption on this processor
        const char * last_enable;
#endif
};

void  ?{}(processor & this, const char name[], struct cluster & cltr);
void ^?{}(processor & this);

static inline void  ?{}(processor & this)                    { this{ "Anonymous Processor", *mainCluster}; }
static inline void  ?{}(processor & this, struct cluster & cltr)    { this{ "Anonymous Processor", cltr}; }
static inline void  ?{}(processor & this, const char name[]) { this{name, *mainCluster }; }

static inline Link(processor) * ?`next( processor * this ) { return &this->link; }

//-----------------------------------------------------------------------------
// I/O
struct __io_data;

#define CFA_CLUSTER_IO_POLLER_USER_THREAD    (1 << 0) // 0x01
#define CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS (1 << 1) // 0x02
#define CFA_CLUSTER_IO_EAGER_SUBMITS         (1 << 2) // 0x04
#define CFA_CLUSTER_IO_KERNEL_POLL_SUBMITS   (1 << 3) // 0x08
#define CFA_CLUSTER_IO_KERNEL_POLL_COMPLETES (1 << 4) // 0x10
#define CFA_CLUSTER_IO_BUFFLEN_OFFSET        16


//-----------------------------------------------------------------------------
// Cluster Tools

// Intrusives lanes which are used by the relaxed ready queue
struct __attribute__((aligned(128))) __intrusive_lane_t;
void  ?{}(__intrusive_lane_t & this);
void ^?{}(__intrusive_lane_t & this);

// Counter used for wether or not the lanes are all empty
struct __attribute__((aligned(128))) __snzi_node_t;
struct __snzi_t {
        unsigned mask;
        int root;
        __snzi_node_t * nodes;
};

void  ?{}( __snzi_t & this, unsigned depth );
void ^?{}( __snzi_t & this );

//TODO adjust cache size to ARCHITECTURE
// Structure holding the relaxed ready queue
struct __ready_queue_t {
        // Data tracking how many/which lanes are used
        // Aligned to 128 for cache locality
        __snzi_t snzi;

        // Data tracking the actual lanes
        // On a seperate cacheline from the used struct since
        // used can change on each push/pop but this data
        // only changes on shrink/grow
        struct {
                // Arary of lanes
                __intrusive_lane_t * volatile data;

                // Number of lanes (empty or not)
                volatile size_t count;
        } lanes;
};

void  ?{}(__ready_queue_t & this);
void ^?{}(__ready_queue_t & this);

//-----------------------------------------------------------------------------
// Cluster
struct __attribute__((aligned(128))) cluster {
        // Ready queue for threads
        __ready_queue_t ready_queue;

        // Name of the cluster
        const char * name;

        // Preemption rate on this cluster
        Duration preemption_rate;

        // List of idle processors
        StackLF(processor) idles;
        volatile unsigned int nprocessors;

        // List of threads
        __spinlock_t thread_list_lock;
        __dllist_t(struct $thread) threads;
        unsigned int nthreads;

        // Link lists fields
        struct __dbg_node_cltr {
                cluster * next;
                cluster * prev;
        } node;

        struct __io_data * io;

        #if !defined(__CFA_NO_STATISTICS__)
                struct __stats_t * stats;
                int print_stats;
        #endif
};
extern Duration default_preemption();

void ?{} (cluster & this, const char name[], Duration preemption_rate, unsigned flags);
void ^?{}(cluster & this);

static inline void ?{} (cluster & this)                                           { this{"Anonymous Cluster", default_preemption(), 0}; }
static inline void ?{} (cluster & this, Duration preemption_rate)                 { this{"Anonymous Cluster", preemption_rate, 0}; }
static inline void ?{} (cluster & this, const char name[])                        { this{name, default_preemption(), 0}; }
static inline void ?{} (cluster & this, unsigned flags)                           { this{"Anonymous Cluster", default_preemption(), flags}; }
static inline void ?{} (cluster & this, Duration preemption_rate, unsigned flags) { this{"Anonymous Cluster", preemption_rate, flags}; }
static inline void ?{} (cluster & this, const char name[], unsigned flags)        { this{name, default_preemption(), flags}; }

static inline [cluster *&, cluster *& ] __get( cluster & this ) __attribute__((const)) { return this.node.[next, prev]; }

static inline struct processor * active_processor() { return TL_GET( this_processor ); } // UNSAFE
static inline struct cluster   * active_cluster  () { return TL_GET( this_processor )->cltr; }

#if !defined(__CFA_NO_STATISTICS__)
        static inline void print_stats_at_exit( cluster & this, int flags ) {
                this.print_stats |= flags;
        }

        static inline void print_stats_at_exit( processor & this, int flags ) {
                this.print_stats |= flags;
        }

        void print_halts( processor & this );
#endif

// Local Variables: //
// mode: c //
// tab-width: 4 //
// End: //